Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method, implemented on a machine having at least one processor, storage, and a communication platform connected to a network for providing a personalized snippet, comprising: receiving, via the communication platform, a request for a snippet related to content to be provided to a user; obtaining a plurality of portions of the content; calculating, for each of the plurality of portions, a first score based on information about of the user; extracting one or more first features from the content; obtaining one or more second features of each of the plurality of portions; comparing, for each of the plurality of portions, the one or more first features with the one or more second features to calculate a second score representing a descriptive power of the portion with respect to the content; generating a first ranked list of the plurality of portions based on the second score for each of the plurality of portions; generating a second ranked list based on the first ranked list and the first score for each of the plurality of portions; selecting one or more portions from the plurality of portions based on the first ranked list and the second ranked list; creating the snippet related to the content based on the one or more portions by modifying the one or more portions based on information related to interaction activity of other users; and generating, in response to the request, a webpage including the snippet.
This invention relates to personalized content summarization in digital systems. The problem addressed is the need to generate concise, relevant snippets of content tailored to individual users while ensuring the snippets accurately represent the original material. The method operates on a machine with processing, storage, and network connectivity. When a user requests a snippet related to specific content, the system retrieves multiple segments of that content. Each segment is evaluated using two scoring mechanisms. The first score assesses relevance to the user based on their profile or behavior data. The second score measures how well each segment describes the entire content by comparing extracted features of the full content against features of the segment. The segments are then ranked twice: first by descriptive power, then by user relevance. The system selects segments from both rankings to create a snippet, which is further refined by analyzing interaction patterns of other users to optimize presentation. Finally, the personalized snippet is embedded in a webpage and delivered to the user. This approach ensures snippets are both user-specific and representative of the original content.
2. The method of claim 1 , wherein the first score represents a likelihood that the user will follow a reference to the content if the snippet includes the portion.
This invention relates to content recommendation systems that analyze user engagement with content snippets to improve recommendation accuracy. The problem addressed is determining whether to include specific portions of content in a snippet to maximize the likelihood that a user will follow a reference to the full content. The method involves generating a first score representing the probability that a user will follow a reference to the content if a snippet includes a particular portion of the content. This score is derived from historical user interaction data, such as click-through rates or engagement metrics, associated with similar content portions. The system also generates a second score representing the likelihood that the user will follow the reference if the snippet does not include the portion. The method then compares these scores to decide whether to include the portion in the snippet. If the first score is higher, the portion is included; otherwise, it is omitted. This approach optimizes snippet composition to enhance user engagement and improve the effectiveness of content recommendations. The invention may be applied in search engines, social media platforms, or any system where content snippets are used to drive user interaction with full content.
3. The method of claim 2 , wherein the likelihood is calculated based on at least one of: characteristics of the user, characteristics of the content, and characteristics of the plurality of portions.
This invention relates to systems for dynamically selecting and presenting content portions to users based on calculated likelihoods of user engagement. The technology addresses the problem of efficiently delivering relevant content to users in a personalized manner, improving user experience and engagement metrics. The method involves analyzing multiple portions of content to determine their suitability for presentation to a specific user. The likelihood of user engagement with each portion is calculated using at least one of three factors: user characteristics (e.g., preferences, behavior, demographics), content characteristics (e.g., topic, format, length), and characteristics of the content portions themselves (e.g., relevance, recency, popularity). These factors are processed to generate a likelihood score for each portion, which is then used to select and present the most engaging content to the user. The method may also involve tracking user interactions with presented content to refine future selections, ensuring continuous improvement in personalization. By dynamically adjusting content presentation based on real-time data, the system enhances user engagement and satisfaction. This approach is particularly useful in applications like social media, news feeds, and recommendation systems where personalized content delivery is critical.
4. The method of claim 2 , wherein transmitting further comprises: providing the reference to the content with the snippet to the user.
This invention describes a method where a machine generates a personalized snippet for content requested by a user. It achieves this by breaking content into portions, then calculating two types of scores for each portion: a "first score" reflecting how likely the user is to click a link to the main content if that portion is in the snippet, and a "second score" measuring how well the portion describes the overall content. These scores, combined with modifications based on other users' interaction activity, help select and assemble the final snippet. When this personalized snippet is generated as part of a webpage to be shown to the user, the system further includes and displays a direct reference (a link) to the full content alongside the snippet, making it convenient for the user to access the complete article if the tailored snippet captures their interest. ERROR (embedding): Error: Failed to save embedding: Could not find the 'embedding' column of 'patent_claims' in the schema cache
5. The method of claim 1 , wherein the one or more portions are associated with highest first scores among the first ranked list of the plurality of portions.
A system and method for ranking and selecting portions of data, such as text segments, based on relevance or importance. The invention addresses the challenge of efficiently identifying the most significant portions from a larger dataset, which is critical in applications like document summarization, information retrieval, or machine learning feature selection. The method involves generating a ranked list of portions from a dataset, where each portion is assigned a first score indicating its relevance or importance. The invention then selects one or more portions from this ranked list, specifically those associated with the highest first scores. This ensures that the most relevant or important portions are prioritized for further processing or output. The selection process may involve additional criteria or thresholds to refine the chosen portions. The method is applicable in various domains, including natural language processing, data analysis, and automated content generation, where identifying key segments of data is essential for accuracy and efficiency.
6. The method of claim 1 , wherein obtaining the plurality of portions of the content includes parsing the content into a plurality of text portions based on a topic of each text portion.
This invention relates to content processing systems that analyze and segment textual data. The problem addressed is the need to efficiently parse large volumes of text into meaningful portions for further analysis, such as topic modeling, summarization, or classification. Traditional methods often rely on fixed-length segmentation or simple keyword-based splitting, which may not accurately capture semantic boundaries between topics. The invention describes a method for dividing content into multiple text portions based on topic coherence. The content is first analyzed to identify distinct topics within the text. Each portion is then extracted such that the text within a single portion pertains to a single topic, ensuring semantic consistency. This approach improves the accuracy of downstream tasks by maintaining contextual relevance within each segment. The method may also include preprocessing steps like removing noise or normalizing text to enhance topic detection. The segmented portions can then be used for further processing, such as machine learning training, document summarization, or information retrieval. This technique is particularly useful in applications requiring high-precision text analysis, such as legal document review, academic research, or automated content moderation.
7. The method of claim 1 , wherein calculating the first score comprises: obtaining the user information comprising a user profile and/or a user interest profile of the user; and determining one or more interests of the user based on the user profile and/or the user interest profile, wherein the first score is calculated based on the one or more second features and the one or more interests of the user.
Online content recommendation systems. Problem: improving the accuracy of recommending content to users. A method for calculating a first score related to user content preference. This calculation involves obtaining user information, which includes a user profile and/or a user interest profile. Based on this user profile and/or user interest profile, one or more interests of the user are determined. The first score is then computed using these determined user interests and one or more second features. These second features are not explicitly defined in this claim but are utilized in the score calculation alongside the user's interests.
8. The method of claim 1 , wherein each of the one or more first features and the one or more second features includes at least one of: length, position, similarity to a title, containment of name entities, and keywords or categories of the content.
This invention relates to a method for analyzing and categorizing content based on extracted features. The method addresses the challenge of efficiently processing and organizing large volumes of textual or multimedia content by identifying and utilizing specific characteristics of the content to improve classification, retrieval, and analysis. The method involves extracting one or more first features and one or more second features from the content. These features are used to determine the relevance, categorization, or contextual meaning of the content. The first and second features may include various attributes such as length, position within the content, similarity to a title, presence of named entities, and keywords or categories associated with the content. Named entities refer to identifiable elements such as people, organizations, locations, or dates, while keywords and categories help in classifying the content into predefined groups. By analyzing these features, the method enables more accurate and efficient content processing, improving tasks such as information retrieval, content filtering, and automated summarization. The approach enhances the ability to extract meaningful insights from unstructured data, making it valuable for applications in search engines, content management systems, and data analytics platforms. The method ensures that the extracted features are comprehensive and diverse, allowing for robust content analysis across different domains.
9. The method of claim 1 , wherein creating the snippet comprises: creating the snippet based on the one or more modified portions based on grammar information.
This invention relates to natural language processing and text summarization, specifically improving the generation of text snippets from source documents. The problem addressed is the creation of coherent and grammatically correct snippets when extracting or modifying portions of text, ensuring the output remains readable and contextually accurate. The method involves generating a snippet from one or more modified portions of a source text. The key improvement is that the snippet creation process incorporates grammar information to refine the extracted or modified text. This ensures that the resulting snippet adheres to grammatical rules, maintains proper syntax, and remains coherent. The grammar information may include part-of-speech tags, syntactic structures, or other linguistic rules that guide the transformation of the modified portions into a well-formed snippet. The modified portions of the text may be derived from operations such as paraphrasing, rephrasing, or extracting segments from the original document. By applying grammar information during snippet creation, the method avoids unnatural or grammatically incorrect outputs, enhancing the quality and usability of the generated snippets. This approach is particularly useful in applications like automated summarization, content generation, and text editing, where maintaining linguistic correctness is critical.
10. The method of claim 1 , wherein calculating the first score comprises using a supervised learning model, such that the first ranked list is generated further based on the supervised learning model.
This invention relates to a method for generating a ranked list of items using a supervised learning model. The method addresses the challenge of efficiently ranking items based on relevance or importance, particularly in scenarios where traditional ranking algorithms may not account for complex patterns or relationships within the data. The method involves calculating a first score for each item in a dataset, where the score is determined using a supervised learning model. This model is trained on labeled data to learn relationships between item features and desired rankings. The first ranked list is then generated based on these scores, ensuring that the ranking reflects the learned patterns from the training data. The supervised learning model may incorporate various machine learning techniques, such as regression, classification, or neural networks, to improve the accuracy and relevance of the rankings. By leveraging supervised learning, the method enhances the ability to rank items in a way that aligns with specific objectives or user preferences, making it particularly useful in applications like recommendation systems, search engines, or data prioritization tasks. The method ensures that the ranked list is dynamically adjusted based on the model's predictions, improving adaptability to changing data or requirements.
11. The method of claim 1 , wherein generating the second ranked list comprises: determining, for each portion of the plurality of portions of the first ranked list, whether a corresponding first score for each portion exceeds a threshold, wherein the one or more portions are selected based on the corresponding first score exceeding the threshold.
This invention relates to ranking and filtering data portions within a ranked list to improve relevance or efficiency. The method addresses the problem of managing large datasets where initial rankings may include irrelevant or low-quality portions, requiring further refinement. The process begins with a first ranked list containing multiple portions, each assigned a first score indicating relevance or quality. The method then evaluates each portion by comparing its first score to a predefined threshold. Only portions with scores exceeding this threshold are selected for inclusion in a second ranked list. This filtering step ensures that the final output contains only the most relevant or highest-quality portions, improving the efficiency and accuracy of subsequent analysis or decision-making. The threshold can be dynamically adjusted based on system requirements or user preferences, allowing flexibility in balancing between strict filtering and broader inclusion. This approach is particularly useful in applications like search engines, recommendation systems, or data analysis tools where initial rankings may need refinement to enhance performance. By focusing on portions that meet a minimum quality standard, the method optimizes resource usage and improves the reliability of the ranked results.
12. The method of claim 1 , further comprising: obtaining, based on a user ID associated with the user, a user profile for the user from a first database; obtaining, based on the user ID, a user interest profile for the user from a second database; and determining the information about the first user based on the user profile and the user interest profile.
This invention relates to personalized information retrieval systems that enhance user experience by integrating multiple user data sources. The problem addressed is the lack of comprehensive user profiling in existing systems, which often rely on limited or fragmented data, leading to less relevant or personalized information delivery. The method involves obtaining a user profile from a first database and a user interest profile from a second database, both based on a user ID. The user profile contains general information about the user, such as demographic data, preferences, or historical interactions. The user interest profile captures more dynamic or specific interests, such as recent activities, search queries, or engagement patterns. By combining these profiles, the system determines more accurate and contextually relevant information about the user. This integrated approach ensures that recommendations, search results, or content suggestions are tailored to both the user's static attributes and evolving interests, improving relevance and engagement. The method may also involve analyzing the user's behavior over time to refine the profiles dynamically, ensuring the system adapts to changing user preferences. This enhances the accuracy of personalized information delivery, making it more effective in meeting user needs. The system can be applied in various domains, including e-commerce, social media, or content recommendation platforms, where personalized user experiences are critical.
13. The method of claim 1 , wherein the interaction activity is one of an online interaction activity and offline interaction activity.
This invention relates to systems and methods for tracking and analyzing interaction activities, both online and offline, to improve user engagement and system performance. The core technology involves monitoring various types of user interactions with a system, such as clicks, messages, transactions, or physical actions, to gather data that can be processed to enhance functionality, personalization, or security. The method includes detecting and categorizing interaction activities, which can occur either online (e.g., digital transactions, social media engagement) or offline (e.g., in-store purchases, physical device usage). These activities are logged and analyzed to derive insights, such as user behavior patterns, system performance metrics, or security threats. The collected data may be used to optimize system responses, tailor user experiences, or detect anomalies. The system may include sensors, software agents, or other monitoring tools to capture interaction data in real time or near real time. The data is then processed using algorithms to classify, correlate, or predict outcomes based on the interactions. For example, online interactions might involve tracking website navigation, while offline interactions could include monitoring physical device usage or in-person transactions. The invention aims to provide a comprehensive approach to interaction tracking, enabling systems to adapt dynamically to user needs, improve efficiency, and enhance security by distinguishing between different types of activities. This method ensures that both digital and physical interactions are accounted for, allowing for a more holistic analysis of user behavior and system performance.
14. A system, having at least one processor, storage, and a communication platform connected to a network for providing a personalized snippet, comprising: a snippet request analyzer configured to receive, via the communication platform, a request for a snippet related to content to be provided to a user; a content parsing unit configured to obtain a plurality of portions of the content; a user interest-based text ranking unit configured to calculate, for each of the plurality of portions, a first score based on information about the user; a content retriever configured to extract one or more first features from the content; a relevance-based text ranking unit configured to: obtain one or more second features of each of the plurality of portions, compare, for each of the plurality of portions, the one or more first features with the one or more second features to calculate a second score representing a descriptive power of the portion with respect to the content, and generate a first ranked list of the plurality of portions based on the second score for each of the plurality of portions; a user interest-based text ranking unit configured to generate a second ranked list based on the first ranked list and the first score for each of the plurality of portions; a snippet generation unit configured to: select one or more portions from the plurality of portions based on the first ranked list and the second ranked list, and create the snippet related to the content based on the selected one or more portions by modifying the one or more portions based on information related to interaction activity of other users; and a snippet transmitting unit configured to generate, in response to the request, a webpage including the snippet.
This system provides personalized content snippets by analyzing user interests and content relevance. The system operates in the domain of content summarization and personalization, addressing the challenge of delivering concise, relevant previews of content tailored to individual users. It processes a request for a snippet related to specific content, then parses the content into multiple portions. A user interest-based ranking unit assigns a score to each portion based on user-specific data, while a content retriever extracts key features from the full content. A relevance-based ranking unit compares these features with each portion to assess descriptive power, generating a ranked list of portions by relevance. The system then combines this relevance ranking with the user interest scores to produce a final ranked list. A snippet generation unit selects portions from this list and modifies them based on interaction patterns of other users, creating a personalized snippet. Finally, the system transmits the snippet as part of a webpage in response to the request. The approach ensures snippets are both relevant to the content and aligned with the user's interests, enhancing engagement and usability.
15. The system of claim 14 , wherein the first score represents a likelihood that the user will follow a reference to the content if the snippet includes the portion.
The system relates to content recommendation and user engagement analysis, specifically improving the presentation of content snippets to users. The problem addressed is determining the optimal way to display portions of content to maximize user engagement, such as clicks or follows, when presenting snippets of larger content items. The system evaluates different portions of content to predict which will most effectively encourage user interaction. The system includes a scoring mechanism that assigns a first score to a portion of content, representing the likelihood that a user will follow a reference to the full content if the snippet includes that portion. This score is derived from historical user interaction data, analyzing how users respond to different content portions in snippets. The system may also generate a second score for the same portion, representing the likelihood that the user will follow the reference if the snippet does not include that portion. By comparing these scores, the system determines the most effective way to present the snippet to maximize engagement. The system further includes a display module that presents the content snippet to the user based on the scores, ensuring that the most engaging portions are included. The scoring mechanism may use machine learning models trained on past user behavior to predict future interactions accurately. The system may also adapt over time, refining its scoring based on new interaction data to improve recommendation accuracy. This approach enhances user experience by delivering more relevant and engaging content snippets, increasing the likelihood of user interaction with the full content.
16. The system of claim 15 , wherein the likelihood is calculated based on at least one of: characteristics of the user, characteristics of the content, and characteristics of the plurality of portions.
This invention relates to a system for dynamically selecting and presenting content portions to a user based on calculated likelihoods of user engagement. The system addresses the challenge of efficiently delivering relevant content to users in a personalized manner, improving user experience and engagement metrics. The system includes a content processing module that divides content into multiple portions, each representing a segment of the original material. A user interaction module tracks user behavior, such as clicks, dwell time, or navigation patterns, to assess engagement with different content portions. A likelihood calculation module computes the probability that a user will engage with a specific content portion, considering factors like user preferences, content relevance, and portion characteristics (e.g., length, topic, or format). The system then selects and presents the most engaging portions to the user, optimizing content delivery based on real-time data. The likelihood calculation may incorporate user characteristics (e.g., demographics, past interactions), content characteristics (e.g., topic, complexity), and portion characteristics (e.g., structure, multimedia elements). By dynamically adjusting content presentation, the system enhances user engagement and reduces the likelihood of disengagement or abandonment. This approach is particularly useful in digital platforms, educational tools, or personalized media services where tailored content delivery is critical.
17. The system of claim 14 , wherein the one or more portions are associated with highest first scores among the first ranked list of the plurality of portions.
A system for processing and analyzing data segments involves ranking portions of data based on relevance or importance. The system generates a first ranked list of multiple data portions, where each portion is assigned a first score indicating its significance. The system then selects one or more portions from this ranked list, specifically those with the highest first scores, for further processing or output. This selection ensures that the most relevant or valuable data portions are prioritized. The system may also generate a second ranked list of the selected portions, where each is assigned a second score, and further refine the selection based on these second scores. The overall process allows for efficient filtering and prioritization of data, improving accuracy and reducing computational overhead in applications such as search engines, recommendation systems, or data analysis tools. The system dynamically adjusts rankings and selections to adapt to changing data or user preferences, ensuring optimal performance in real-time or batch processing scenarios.
18. The system of claim 14 , wherein the first score being calculated comprises the user interest-based text ranking unit being further configured to: obtain the user information comprising a user profile and/or a user interest profile of the user; and determine one or more interests of the user based on the user profile and/or the user interest profile, wherein the first score is calculated based on the one or more second features and the one or more interests of the user.
This invention relates to a system for ranking text content based on user interests to improve relevance in information retrieval or recommendation systems. The problem addressed is the need to personalize content ranking by incorporating user-specific interests alongside traditional text features, enhancing the accuracy and relevance of recommendations or search results. The system includes a user interest-based text ranking unit that calculates a first score for text content by analyzing user information, such as a user profile or a user interest profile. The unit extracts one or more interests of the user from this information. The first score is then computed by combining these user interests with one or more second features derived from the text content itself. This approach ensures that the ranking process accounts for both the inherent qualities of the text and the user's specific preferences, leading to more personalized and relevant results. The system may also include a text feature extraction unit that generates the second features from the text content, such as keywords, semantic embeddings, or other linguistic characteristics. These features are used alongside the user interests to refine the ranking process. By integrating user-specific data with content-based features, the system provides a more dynamic and adaptive ranking mechanism, improving user engagement and satisfaction in applications like search engines, recommendation systems, or content personalization platforms.
19. A machine-readable tangible and non-transitory medium having information recorded thereon for providing a personalized snippet of content, wherein the information, when read by the machine, causes the machine to perform the following: receiving, via the communication platform, a request for a snippet related to content to be provided to a user; obtaining a plurality of portions of the content; calculating, for each of the plurality of portions, a first score based on information about of the user; extracting one or more first features from the content; obtaining one or more second features of each of the plurality of portions; comparing, for each of the plurality of portions, the one or more first features with the one or more second features to calculate a second score representing a descriptive power of the portion with respect to the content; generating a first ranked list of the plurality of portions based on the second score for each of the plurality of portions; generating a second ranked list based on the first ranked list and the first score for each of the plurality of portions; selecting one or more portions from the plurality of portions based on the first ranked list and the second ranked list; creating the snippet related to the content based on the one or more portions by modifying the one or more portions based on information related to interaction activity of other users; and generating, in response to the request, a webpage including the snippet.
This invention relates to personalized content summarization, specifically generating tailored snippets of content for users based on their preferences and interaction patterns. The system addresses the challenge of delivering concise, relevant content excerpts that align with individual user interests while maintaining descriptive accuracy. The process begins by receiving a request for a content snippet via a communication platform. The system then obtains multiple portions of the content and calculates a first score for each portion based on user-specific information, such as preferences or past behavior. Additionally, the system extracts general features from the full content and compares these with features of each portion to compute a second score, which measures how well each portion represents the overall content. A first ranked list is generated based on these second scores, prioritizing portions that best describe the content. A second ranked list is then created by combining the first ranked list with the user-specific first scores, balancing descriptive power with personal relevance. The system selects one or more portions from these ranked lists and modifies them based on interaction data from other users, such as engagement metrics or feedback. Finally, a webpage is generated that includes the personalized snippet, ensuring the content is both concise and tailored to the user's interests. This approach enhances content delivery by dynamically adapting snippets to individual users while preserving the core meaning of the original material.
20. The medium of claim 19 , wherein the first score represents a likelihood that the user will follow a reference to the content if the snippet includes the portion.
This invention relates to systems for selecting and displaying content snippets to users, particularly in digital environments where content recommendations are presented. The problem addressed is improving the relevance and effectiveness of content snippets by dynamically selecting portions of content that maximize user engagement. The invention involves analyzing content to identify portions likely to capture user interest and generating scores that predict the likelihood of user interaction with those portions. These scores are used to prioritize and display the most engaging snippets, thereby increasing the probability that users will follow references to the full content. The system may also compare different portions of content to determine which is most effective for a given user or context, ensuring optimal presentation. By dynamically adjusting snippet selection based on predicted user behavior, the invention enhances content discovery and engagement in digital platforms.
21. The medium of claim 20 , wherein the likelihood is calculated based on at least one of: characteristics of the user, characteristics of the content, and characteristics of the plurality of portions.
This invention relates to systems for selecting and presenting content portions to users based on calculated likelihoods of user engagement. The technology addresses the problem of efficiently delivering relevant content to users in a way that maximizes engagement while minimizing computational overhead. The system analyzes multiple factors to determine the likelihood that a user will engage with specific content portions, such as user characteristics (e.g., preferences, behavior history), content characteristics (e.g., topic, format), and portion characteristics (e.g., length, position within the content). By evaluating these factors, the system dynamically selects and presents the most engaging content portions to the user. The invention also includes mechanisms for adjusting the selection process based on real-time feedback, ensuring continuous optimization of content delivery. This approach improves user experience by providing personalized and relevant content while reducing the computational resources required for content processing and presentation. The system is particularly useful in applications where large volumes of content must be filtered and presented to users in an efficient and engaging manner.
Unknown
April 14, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.